How AI can strengthen energy poverty governance

The researchers also argue that explainable AI models are essential for ensuring fairness and accountability in policy design. In traditional statistical models, the relationships between variables may be difficult to interpret. In contrast, SHAP-based interpretation clearly reveals the magnitude and direction of influence for key features, helping policymakers understand why certain households are consistently at risk and how specific reforms can shift vulnerability profiles over time.

Energy poverty remains one of the most persistent global challenges in the transition toward sustainable development. Policymakers are seeking new tools to understand the structural forces driving vulnerability. Recent advances in explainable AI now offer the ability to identify, predict, and interpret the socioeconomic patterns behind energy insecurity with greater clarity than traditional analytical methods.

A new peer-reviewed article, “Leveraging Explainable AI to Decode Energy Poverty in China: Implications for SDGs and National Policy,” published in Sustainability, analyzes detailed household-level data to examine how energy poverty has evolved across China between 2014 and 2020.

The study presents a novel machine-learning model designed to identify, predict, and interpret the factors that most strongly influence whether a household becomes energy poor. The study is based on extensive micro-level household data to quantify the changing patterns of energy poverty over a six-year period and evaluate how national reforms have reshaped the energy affordability landscape.

AI model achieves high accuracy in detecting energy poverty across China

To assess energy poverty more precisely, the researchers developed the Energy Poverty Prediction and Explanation Framework for China’s Sustainability (EPPE-FCS), a predictive model that uses a Convolutional Neural Network integrated with SHAP explainability techniques. The model was trained on data from the China Family Panel Studies covering the years 2014, 2016, 2018, and 2020, with Spearman-based feature selection used to identify variables most relevant to energy poverty outcomes.

The model achieved an average prediction accuracy of 98.23 percent, outperforming all traditional baseline models, including XGBoost, Random Forest, Support Vector Machine, and Logistic Regression. This high level of precision demonstrates the suitability of deep learning for capturing nonlinear relationships between expenditure patterns, energy use behaviors, and household socioeconomic characteristics.

By embedding explainability into the predictive framework, the study addresses one of the major policy challenges associated with AI applications: the ability to justify model outputs and ensure transparency. The SHAP-based interpretability module provides detailed insights into how each feature contributes to classification outcomes, enabling decision-makers to identify the strongest levers for intervention and monitor trends in energy affordability with higher accountability.

The analysis revealed that per capita household expenditure is the most influential predictor of energy poverty in every year studied. Rather than energy infrastructure limitations or fuel choices alone, the dominant constraint appears to be the economic resources available to households. The model shows that low spending capacity consistently aligns with elevated vulnerability and reduced flexibility in meeting energy needs.

Next, the model identified energy burden indicators, specifically the household shares of gas, electricity, and heating expenditures, as critical determinants. These indicators capture the extent to which essential energy costs consume a household’s budget. Households experiencing high energy burdens are significantly more likely to fall into energy poverty because rising expenditures limit the ability to allocate income to other necessities.

The study’s predictive insights underscore that energy poverty in China is both an economic and structural issue. As energy markets evolve and energy prices shift, the burden on poor households intensifies unless targeted support mechanisms are in place.

Policy reforms reduce gas burden but uneven vulnerabilities persist

The study sheds light on the shifting influence of gas expenditure burden over time. Between 2014 and 2020, the contribution of gas burden to energy poverty decreased steadily. The researchers attribute this trend to several large-scale national reforms, including the Clean Heating Initiative, rural clean energy transition programs, grid modernization, and the expansion of subsidy schemes for low-income households.

These reforms increased access to natural gas and clean heating systems, improved heating efficiency, and helped stabilize fuel costs in regions historically dependent on coal or biomass. The decline in gas burden demonstrates that structural reforms can meaningfully reduce energy-related inequalities when combined with financial support tools.

Despite this progress, the momentum has not fully translated into relief for households bearing heavy electricity and heating burdens. The model shows that for many families, electricity remains a disproportionately high expenditure relative to income, especially in rural and peri-urban regions where seasonal temperature swings increase heating and cooling demands. The data reveals that as access to modern energy improves, consumption increases, but for low-income households, this growth exacerbates affordability pressures.

The AI analysis further identifies a consistent group of vulnerable households across multiple survey waves, indicating that energy poverty persists in cyclical patterns. These households tend to be characterized by low income, restricted expenditure capacity, and high shares of their budgets devoted to essential energy consumption. Such patterns suggest that energy poverty is embedded in broader socioeconomic constraints rather than isolated energy-sector inefficiencies.

The study also highlights significant regional disparities. Rural households are more likely to face higher energy burdens due to colder climates in northern provinces, reliance on older heating systems, and higher per-unit electricity prices in some localities. Urban households, while benefiting from more advanced infrastructure, continue to face affordability challenges stemming from rising electricity consumption and increasing reliance on electric heating and cooling.

The findings show that energy poverty in China is multifaceted, influenced by geography, market structures, climate conditions, and social welfare systems. The authors argue that sustainable development goals cannot be met without addressing the intertwined economic and energy challenges that hold poor households at a structural disadvantage.

Explainable AI offers roadmap for targeted poverty alleviation and SDG progress

The study offers several policy recommendations that highlight the importance of integrating explainable AI into poverty governance frameworks. The EPPE-FCS model offers a pathway for national and regional policymakers to diagnose energy poverty with greater precision and develop interventions aligned with China’s long-term environmental and socioeconomic goals.

The authors recommend a dual-track approach to alleviating energy poverty:

1. Short-term interventions should focus on targeted subsidies that reduce the immediate financial pressure on low-income households. By lowering effective energy prices or providing direct cash transfers, governments can help vulnerable families maintain safe and adequate energy consumption levels without sacrificing other basic needs.

2. Long-term strategies should prioritize structural improvements, including income enhancement programs, expansion of clean and efficient energy services, promotion of energy-efficient appliances, and modernization of rural heating systems. These measures aim to reduce reliance on costly or inefficient fuels and create a more equitable distribution of energy services.

These interventions support multiple Sustainable Development Goals, particularly SDG 7 on affordable and clean energy, SDG 1 on poverty reduction, and SDG 13 on climate action. By enabling transparent, evidence-based targeting of policy measures, explainable AI tools can help China advance toward its national goals for carbon neutrality and sustainable energy transition.

The researchers also argue that explainable AI models are essential for ensuring fairness and accountability in policy design. In traditional statistical models, the relationships between variables may be difficult to interpret. In contrast, SHAP-based interpretation clearly reveals the magnitude and direction of influence for key features, helping policymakers understand why certain households are consistently at risk and how specific reforms can shift vulnerability profiles over time.